[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

altivec kernels


I include below all of my AltiVec enhanced gemm kernels.  You'll need to
move the include atlas_prefetch.h into your ATLAS/include directory.  
Everything else goes in ATLAS/tune/blas/gemm/CASES.  Results are not too
spectacular: the full SGEMM seems to peak at just under 1.9Gflop, and
the full DGEMM noses up to around 670Mflop, all on a 533Mhz G4.  Complex
are roughly the same as their real counterparts.

This is pretty much all I'm gonna do for now, since I need to get working
on the next developer release that supports altivec before improving the
kernels is very helpful.